0fb9afd3c8aa55717b86e8fa7ea445fb937f82f2
[thirdparty/mdadm.git] / Incremental.c
1 /*
2 * Incremental.c - support --incremental. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
4 *
5 * Copyright (C) 2006 Neil Brown <neilb@suse.de>
6 *
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 *
22 * Author: Neil Brown
23 * Email: <neilb@suse.de>
24 * Paper: Neil Brown
25 * Novell Inc
26 * GPO Box Q1283
27 * QVB Post Office, NSW 1230
28 * Australia
29 */
30
31 #include "mdadm.h"
32
33 static int count_active(struct supertype *st, int mdfd, char **availp,
34 struct mdinfo *info);
35 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
36 int number, __u64 events, int verbose,
37 char *array_name);
38
39 int Incremental(char *devname, int verbose, int runstop,
40 struct supertype *st, char *homehost, int autof)
41 {
42 /* Add this device to an array, creating the array if necessary
43 * and starting the array if sensibe or - if runstop>0 - if possible.
44 *
45 * This has several steps:
46 *
47 * 1/ Check if device is permitted by mdadm.conf, reject if not.
48 * 2/ Find metadata, reject if none appropriate (check
49 * version/name from args)
50 * 3/ Check if there is a match in mdadm.conf
51 * 3a/ if not, check for homehost match. If no match, reject.
52 * 4/ Determine device number.
53 * - If in mdadm.conf with std name, use that
54 * - UUID in /var/run/mdadm.map use that
55 * - If name is suggestive, use that. unless in use with different uuid.
56 * - Choose a free, high number.
57 * - Use a partitioned device unless strong suggestion not to.
58 * e.g. auto=md
59 * 5/ Find out if array already exists
60 * 5a/ if it does not
61 * - choose a name, from mdadm.conf or 'name' field in array.
62 * - create the array
63 * - add the device
64 * 5b/ if it does
65 * - check one drive in array to make sure metadata is a reasonably
66 * close match. Reject if not (e.g. different type)
67 * - add the device
68 * 6/ Make sure /var/run/mdadm.map contains this array.
69 * 7/ Is there enough devices to possibly start the array?
70 * 7a/ if not, finish with success.
71 * 7b/ if yes,
72 * - read all metadata and arrange devices like -A does
73 * - if number of OK devices match expected, or -R and there are enough,
74 * start the array (auto-readonly).
75 */
76 struct stat stb;
77 struct mdinfo info, info2;
78 struct mddev_ident_s *array_list, *match;
79 char chosen_name[1024];
80 int rv;
81 int devnum;
82 struct map_ent *mp, *map = NULL;
83 int dfd, mdfd;
84 char *avail;
85 int active_disks;
86
87
88 struct createinfo *ci = conf_get_create_info();
89
90 if (autof == 0)
91 autof = ci->autof;
92
93 /* 1/ Check if devices is permitted by mdadm.conf */
94
95 if (!conf_test_dev(devname)) {
96 if (verbose >= 0)
97 fprintf(stderr, Name
98 ": %s not permitted by mdadm.conf.\n",
99 devname);
100 return 1;
101 }
102
103 /* 2/ Find metadata, reject if none appropriate (check
104 * version/name from args) */
105
106 dfd = dev_open(devname, O_RDONLY|O_EXCL);
107 if (dfd < 0) {
108 if (verbose >= 0)
109 fprintf(stderr, Name ": cannot open %s: %s.\n",
110 devname, strerror(errno));
111 return 1;
112 }
113 if (fstat(dfd, &stb) < 0) {
114 if (verbose >= 0)
115 fprintf(stderr, Name ": fstat failed for %s: %s.\n",
116 devname, strerror(errno));
117 close(dfd);
118 return 1;
119 }
120 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
121 if (verbose >= 0)
122 fprintf(stderr, Name ": %s is not a block device.\n",
123 devname);
124 close(dfd);
125 return 1;
126 }
127
128 if (st == NULL && (st = guess_super(dfd)) == NULL) {
129 if (verbose >= 0)
130 fprintf(stderr, Name
131 ": no recognisable superblock on %s.\n",
132 devname);
133 close(dfd);
134 return 1;
135 }
136 if (st->ss->load_super(st, dfd, NULL)) {
137 if (verbose >= 0)
138 fprintf(stderr, Name ": no RAID superblock on %s.\n",
139 devname);
140 close(dfd);
141 return 1;
142 }
143 st->ss->getinfo_super(st, &info);
144 close (dfd);
145
146 /* 3/ Check if there is a match in mdadm.conf */
147
148 array_list = conf_get_ident(NULL);
149 match = NULL;
150 for (; array_list; array_list = array_list->next) {
151 if (array_list->uuid_set &&
152 same_uuid(array_list->uuid, info.uuid, st->ss->swapuuid)
153 == 0) {
154 if (verbose >= 2)
155 fprintf(stderr, Name
156 ": UUID differs from %s.\n",
157 array_list->devname);
158 continue;
159 }
160 if (array_list->name[0] &&
161 strcasecmp(array_list->name, info.name) != 0) {
162 if (verbose >= 2)
163 fprintf(stderr, Name
164 ": Name differs from %s.\n",
165 array_list->devname);
166 continue;
167 }
168 if (array_list->devices &&
169 !match_oneof(array_list->devices, devname)) {
170 if (verbose >= 2)
171 fprintf(stderr, Name
172 ": Not a listed device for %s.\n",
173 array_list->devname);
174 continue;
175 }
176 if (array_list->super_minor != UnSet &&
177 array_list->super_minor != info.array.md_minor) {
178 if (verbose >= 2)
179 fprintf(stderr, Name
180 ": Different super-minor to %s.\n",
181 array_list->devname);
182 continue;
183 }
184 if (!array_list->uuid_set &&
185 !array_list->name[0] &&
186 !array_list->devices &&
187 array_list->super_minor == UnSet) {
188 if (verbose >= 2)
189 fprintf(stderr, Name
190 ": %s doesn't have any identifying information.\n",
191 array_list->devname);
192 continue;
193 }
194 /* FIXME, should I check raid_disks and level too?? */
195
196 if (match) {
197 if (verbose >= 0)
198 fprintf(stderr, Name
199 ": we match both %s and %s - cannot decide which to use.\n",
200 match->devname, array_list->devname);
201 return 2;
202 }
203 match = array_list;
204 }
205
206 /* 3a/ if not, check for homehost match. If no match, reject. */
207 if (!match) {
208 if (homehost == NULL ||
209 st->ss->match_home(st, homehost) == 0) {
210 if (verbose >= 0)
211 fprintf(stderr, Name
212 ": not found in mdadm.conf and not identified by homehost.\n");
213 return 2;
214 }
215 }
216 /* 4/ Determine device number. */
217 /* - If in mdadm.conf with std name, use that */
218 /* - UUID in /var/run/mdadm.map use that */
219 /* - If name is suggestive, use that. unless in use with */
220 /* different uuid. */
221 /* - Choose a free, high number. */
222 /* - Use a partitioned device unless strong suggestion not to. */
223 /* e.g. auto=md */
224 if (match && is_standard(match->devname, &devnum))
225 /* We have devnum now */;
226 else if ((mp = map_by_uuid(&map, info.uuid)) != NULL)
227 devnum = mp->devnum;
228 else {
229 /* Have to guess a bit. */
230 int use_partitions = 1;
231 char *np, *ep;
232 if ((autof&7) == 3 || (autof&7) == 5)
233 use_partitions = 0;
234 np = strchr(info.name, ':');
235 if (np)
236 np++;
237 else
238 np = info.name;
239 devnum = strtoul(np, &ep, 10);
240 if (ep > np && *ep == 0) {
241 /* This is a number. Let check that it is unused. */
242 if (mddev_busy(use_partitions ? (-1-devnum) : devnum))
243 devnum = -1;
244 } else
245 devnum = -1;
246
247 if (devnum < 0) {
248 /* Haven't found anything yet, choose something free */
249 devnum = find_free_devnum(use_partitions);
250
251 if (devnum == NoMdDev) {
252 fprintf(stderr, Name
253 ": No spare md devices!!\n");
254 return 2;
255 }
256 } else
257 devnum = use_partitions ? (-1-devnum) : devnum;
258 }
259 mdfd = open_mddev_devnum(match ? match->devname : NULL,
260 devnum,
261 info.name,
262 chosen_name, autof >> 3);
263 if (mdfd < 0) {
264 fprintf(stderr, Name ": failed to open %s: %s.\n",
265 chosen_name, strerror(errno));
266 return 2;
267 }
268 /* 5/ Find out if array already exists */
269 if (! mddev_busy(devnum)) {
270 /* 5a/ if it does not */
271 /* - choose a name, from mdadm.conf or 'name' field in array. */
272 /* - create the array */
273 /* - add the device */
274 mdu_array_info_t ainf;
275 mdu_disk_info_t disk;
276 char md[20];
277 struct mdinfo *sra;
278
279 memset(&ainf, 0, sizeof(ainf));
280 ainf.major_version = st->ss->major;
281 ainf.minor_version = st->minor_version;
282 if (ioctl(mdfd, SET_ARRAY_INFO, &ainf) != 0) {
283 fprintf(stderr, Name
284 ": SET_ARRAY_INFO failed for %s: %s\b",
285 chosen_name, strerror(errno));
286 close(mdfd);
287 return 2;
288 }
289 sprintf(md, "%d.%d\n", st->ss->major, st->minor_version);
290 sra = sysfs_read(mdfd, devnum, GET_VERSION);
291 sysfs_set_str(sra, NULL, "metadata_version", md);
292 memset(&disk, 0, sizeof(disk));
293 disk.major = major(stb.st_rdev);
294 disk.minor = minor(stb.st_rdev);
295 sysfs_free(sra);
296 if (ioctl(mdfd, ADD_NEW_DISK, &disk) != 0) {
297 fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
298 devname, chosen_name, strerror(errno));
299 ioctl(mdfd, STOP_ARRAY, 0);
300 close(mdfd);
301 return 2;
302 }
303 sra = sysfs_read(mdfd, devnum, GET_DEVS);
304 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
305 /* It really should be 'none' - must be old buggy
306 * kernel, and mdadm -I may not be able to complete.
307 * So reject it.
308 */
309 ioctl(mdfd, STOP_ARRAY, NULL);
310 fprintf(stderr, Name
311 ": You have an old buggy kernel which cannot support\n"
312 " --incremental reliably. Aborting.\n");
313 close(mdfd);
314 sysfs_free(sra);
315 return 2;
316 }
317 } else {
318 /* 5b/ if it does */
319 /* - check one drive in array to make sure metadata is a reasonably */
320 /* close match. Reject if not (e.g. different type) */
321 /* - add the device */
322 char dn[20];
323 int dfd2;
324 mdu_disk_info_t disk;
325 int err;
326 struct mdinfo *sra;
327 struct supertype *st2;
328 sra = sysfs_read(mdfd, devnum, (GET_VERSION | GET_DEVS |
329 GET_STATE));
330
331 if (sra->array.major_version != st->ss->major ||
332 sra->array.minor_version != st->minor_version) {
333 if (verbose >= 0)
334 fprintf(stderr, Name
335 ": %s has different metadata to chosen array %s %d.%d %d.%d.\n",
336 devname, chosen_name,
337 sra->array.major_version,
338 sra->array.minor_version,
339 st->ss->major, st->minor_version);
340 close(mdfd);
341 return 1;
342 }
343 sprintf(dn, "%d:%d", sra->devs->disk.major,
344 sra->devs->disk.minor);
345 dfd2 = dev_open(dn, O_RDONLY);
346 st2 = dup_super(st);
347 if (st2->ss->load_super(st2, dfd2, NULL)) {
348 fprintf(stderr, Name
349 ": Strange error loading metadata for %s.\n",
350 chosen_name);
351 close(mdfd);
352 close(dfd2);
353 return 2;
354 }
355 close(dfd2);
356 st2->ss->getinfo_super(st2, &info2);
357 st2->ss->free_super(st2);
358 if (info.array.level != info2.array.level ||
359 memcmp(info.uuid, info2.uuid, 16) != 0 ||
360 info.array.raid_disks != info2.array.raid_disks) {
361 fprintf(stderr, Name
362 ": unexpected difference between %s and %s.\n",
363 chosen_name, devname);
364 close(mdfd);
365 return 2;
366 }
367 memset(&disk, 0, sizeof(disk));
368 disk.major = major(stb.st_rdev);
369 disk.minor = minor(stb.st_rdev);
370 err = ioctl(mdfd, ADD_NEW_DISK, &disk);
371 if (err < 0 && errno == EBUSY) {
372 /* could be another device present with the same
373 * disk.number. Find and reject any such
374 */
375 find_reject(mdfd, st, sra, info.disk.number,
376 info.events, verbose, chosen_name);
377 err = ioctl(mdfd, ADD_NEW_DISK, &disk);
378 }
379 if (err < 0) {
380 fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
381 devname, chosen_name, strerror(errno));
382 close(mdfd);
383 return 2;
384 }
385 }
386 /* 6/ Make sure /var/run/mdadm.map contains this array. */
387 map_update(&map, devnum,
388 info.array.major_version,
389 info.array.minor_version,
390 info.uuid, chosen_name);
391
392 /* 7/ Is there enough devices to possibly start the array? */
393 /* 7a/ if not, finish with success. */
394 avail = NULL;
395 active_disks = count_active(st, mdfd, &avail, &info);
396 if (enough(info.array.level, info.array.raid_disks,
397 info.array.layout, info.array.state & 1,
398 avail, active_disks) == 0) {
399 free(avail);
400 if (verbose >= 0)
401 fprintf(stderr, Name
402 ": %s attached to %s, not enough to start (%d).\n",
403 devname, chosen_name, active_disks);
404 close(mdfd);
405 return 0;
406 }
407 free(avail);
408
409 /* 7b/ if yes, */
410 /* - if number of OK devices match expected, or -R and there */
411 /* are enough, */
412 /* + add any bitmap file */
413 /* + start the array (auto-readonly). */
414 {
415 mdu_array_info_t ainf;
416
417 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
418 if (verbose >= 0)
419 fprintf(stderr, Name
420 ": %s attached to %s which is already active.\n",
421 devname, chosen_name);
422 close (mdfd);
423 return 0;
424 }
425 }
426 if (runstop > 0 || active_disks >= info.array.working_disks) {
427 struct mdinfo *sra;
428 /* Let's try to start it */
429 if (match && match->bitmap_file) {
430 int bmfd = open(match->bitmap_file, O_RDWR);
431 if (bmfd < 0) {
432 fprintf(stderr, Name
433 ": Could not open bitmap file %s.\n",
434 match->bitmap_file);
435 close(mdfd);
436 return 1;
437 }
438 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
439 close(bmfd);
440 fprintf(stderr, Name
441 ": Failed to set bitmapfile for %s.\n",
442 chosen_name);
443 close(mdfd);
444 return 1;
445 }
446 close(bmfd);
447 }
448 sra = sysfs_read(mdfd, devnum, 0);
449 if (sra == NULL || active_disks >= info.array.working_disks)
450 rv = ioctl(mdfd, RUN_ARRAY, NULL);
451 else
452 rv = sysfs_set_str(sra, NULL,
453 "array_state", "read-auto");
454 if (rv == 0) {
455 if (verbose >= 0)
456 fprintf(stderr, Name
457 ": %s attached to %s, which has been started.\n",
458 devname, chosen_name);
459 rv = 0;
460 } else {
461 fprintf(stderr, Name
462 ": %s attached to %s, but failed to start: %s.\n",
463 devname, chosen_name, strerror(errno));
464 rv = 1;
465 }
466 } else {
467 if (verbose >= 0)
468 fprintf(stderr, Name
469 ": %s attached to %s, not enough to start safely.\n",
470 devname, chosen_name);
471 rv = 0;
472 }
473 close(mdfd);
474 return rv;
475 }
476
477 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
478 int number, __u64 events, int verbose,
479 char *array_name)
480 {
481 /* Find a device attached to this array with a disk.number of number
482 * and events less than the passed events, and remove the device.
483 */
484 struct mdinfo *d;
485 mdu_array_info_t ra;
486
487 if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
488 return; /* not safe to remove from active arrays
489 * without thinking more */
490
491 for (d = sra->devs; d ; d = d->next) {
492 char dn[10];
493 int dfd;
494 struct mdinfo info;
495 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
496 dfd = dev_open(dn, O_RDONLY);
497 if (dfd < 0)
498 continue;
499 if (st->ss->load_super(st, dfd, NULL)) {
500 close(dfd);
501 continue;
502 }
503 st->ss->getinfo_super(st, &info);
504 st->ss->free_super(st);
505 close(dfd);
506
507 if (info.disk.number != number ||
508 info.events >= events)
509 continue;
510
511 if (d->disk.raid_disk > -1)
512 sysfs_set_str(sra, d, "slot", "none");
513 if (sysfs_set_str(sra, d, "state", "remove") == 0)
514 if (verbose >= 0)
515 fprintf(stderr, Name
516 ": removing old device %s from %s\n",
517 d->sys_name+4, array_name);
518 }
519 }
520
521 static int count_active(struct supertype *st, int mdfd, char **availp,
522 struct mdinfo *bestinfo)
523 {
524 /* count how many devices in sra think they are active */
525 struct mdinfo *d;
526 int cnt = 0, cnt1 = 0;
527 __u64 max_events = 0;
528 struct mdinfo *sra = sysfs_read(mdfd, -1, GET_DEVS | GET_STATE);
529 char *avail = NULL;
530
531 for (d = sra->devs ; d ; d = d->next) {
532 char dn[30];
533 int dfd;
534 int ok;
535 struct mdinfo info;
536
537 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
538 dfd = dev_open(dn, O_RDONLY);
539 if (dfd < 0)
540 continue;
541 ok = st->ss->load_super(st, dfd, NULL);
542 close(dfd);
543 if (ok != 0)
544 continue;
545 st->ss->getinfo_super(st, &info);
546 if (info.disk.state & (1<<MD_DISK_SYNC))
547 {
548 if (avail == NULL) {
549 avail = malloc(info.array.raid_disks);
550 memset(avail, 0, info.array.raid_disks);
551 }
552 if (cnt == 0) {
553 cnt++;
554 max_events = info.events;
555 avail[info.disk.raid_disk] = 2;
556 st->ss->getinfo_super(st, bestinfo);
557 } else if (info.events == max_events) {
558 cnt++;
559 avail[info.disk.raid_disk] = 2;
560 } else if (info.events == max_events-1) {
561 cnt1++;
562 avail[info.disk.raid_disk] = 1;
563 } else if (info.events < max_events - 1)
564 ;
565 else if (info.events == max_events+1) {
566 int i;
567 cnt1 = cnt;
568 cnt = 1;
569 max_events = info.events;
570 for (i=0; i<info.array.raid_disks; i++)
571 if (avail[i])
572 avail[i]--;
573 avail[info.disk.raid_disk] = 2;
574 st->ss->getinfo_super(st, bestinfo);
575 } else { /* info.events much bigger */
576 cnt = 1; cnt1 = 0;
577 memset(avail, 0, info.disk.raid_disk);
578 max_events = info.events;
579 st->ss->getinfo_super(st, bestinfo);
580 }
581 }
582 st->ss->free_super(st);
583 }
584 return cnt + cnt1;
585 }
586
587 void RebuildMap(void)
588 {
589 struct mdstat_ent *mdstat = mdstat_read(0, 0);
590 struct mdstat_ent *md;
591 struct map_ent *map = NULL;
592 int mdp = get_mdp_major();
593
594 for (md = mdstat ; md ; md = md->next) {
595 struct mdinfo *sra = sysfs_read(-1, md->devnum, GET_DEVS);
596 struct mdinfo *sd;
597
598 for (sd = sra->devs ; sd ; sd = sd->next) {
599 char dn[30];
600 int dfd;
601 int ok;
602 struct supertype *st;
603 char *path;
604 struct mdinfo info;
605
606 sprintf(dn, "%d:%d", sd->disk.major, sd->disk.minor);
607 dfd = dev_open(dn, O_RDONLY);
608 if (dfd < 0)
609 continue;
610 st = guess_super(dfd);
611 if ( st == NULL)
612 ok = -1;
613 else
614 ok = st->ss->load_super(st, dfd, NULL);
615 close(dfd);
616 if (ok != 0)
617 continue;
618 st->ss->getinfo_super(st, &info);
619 if (md->devnum > 0)
620 path = map_dev(MD_MAJOR, md->devnum, 0);
621 else
622 path = map_dev(mdp, (-1-md->devnum)<< 6, 0);
623 map_add(&map, md->devnum, st->ss->major,
624 st->minor_version,
625 info.uuid, path ? : "/unknown");
626 st->ss->free_super(st);
627 break;
628 }
629 }
630 map_write(map);
631 map_free(map);
632 }
633
634 int IncrementalScan(int verbose)
635 {
636 /* look at every device listed in the 'map' file.
637 * If one is found that is not running then:
638 * look in mdadm.conf for bitmap file.
639 * if one exists, but array has none, add it.
640 * try to start array in auto-readonly mode
641 */
642 struct map_ent *mapl = NULL;
643 struct map_ent *me;
644 mddev_ident_t devs, mddev;
645 int rv = 0;
646
647 map_read(&mapl);
648 devs = conf_get_ident(NULL);
649
650 for (me = mapl ; me ; me = me->next) {
651 char path[1024];
652 mdu_array_info_t array;
653 mdu_bitmap_file_t bmf;
654 struct mdinfo *sra;
655 int mdfd = open_mddev_devnum(me->path, me->devnum,
656 NULL, path, 0);
657 if (mdfd < 0)
658 continue;
659 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
660 errno != ENODEV) {
661 close(mdfd);
662 continue;
663 }
664 /* Ok, we can try this one. Maybe it needs a bitmap */
665 for (mddev = devs ; mddev ; mddev = mddev->next)
666 if (strcmp(mddev->devname, me->path) == 0)
667 break;
668 if (mddev && mddev->bitmap_file) {
669 /*
670 * Note: early kernels will wrongly fail this, so it
671 * is a hint only
672 */
673 int added = -1;
674 if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
675 int bmfd = open(mddev->bitmap_file, O_RDWR);
676 if (bmfd >= 0) {
677 added = ioctl(mdfd, SET_BITMAP_FILE,
678 bmfd);
679 close(bmfd);
680 }
681 }
682 if (verbose >= 0) {
683 if (added == 0)
684 fprintf(stderr, Name
685 ": Added bitmap %s to %s\n",
686 mddev->bitmap_file, me->path);
687 else if (errno != EEXIST)
688 fprintf(stderr, Name
689 ": Failed to add bitmap to %s: %s\n",
690 me->path, strerror(errno));
691 }
692 }
693 sra = sysfs_read(mdfd, 0, 0);
694 if (sra) {
695 if (sysfs_set_str(sra, NULL,
696 "array_state", "read-auto") == 0) {
697 if (verbose >= 0)
698 fprintf(stderr, Name
699 ": started array %s\n",
700 me->path);
701 } else {
702 fprintf(stderr, Name
703 ": failed to start array %s: %s\n",
704 me->path, strerror(errno));
705 rv = 1;
706 }
707 }
708 }
709 return rv;
710 }